docker 容器中未设置 CUDA_HOME 环境变量

Question

我有以下 Dockerfile，尝试创建一个封装 CLIP-Fields github 的容器（https://github.com/notmahi/clip-fields）：

# Use the official PyTorch base image with CUDA support
FROM pytorch/pytorch:1.8.1-cuda11.1-cudnn8-runtime

# Install system dependencies
RUN apt-get update && apt-get install -y git
RUN pip install --upgrade pip
RUN apt-get install g++ -y

# Clone the CLIP Fields repository
RUN git clone https://github.com/notmahi/clip-fields /clip-fields
WORKDIR /clip-fields

# Clone and install the Detic submodule
RUN git clone https://github.com/notmahi/Detic.git

# Clone and install the LSeg submodule
RUN git clone https://github.com/notmahi/LSeg.git

# Create a conda environment and activate it. py 3.8 used
RUN conda create -n cf python=3.8 && \ 
    echo "conda activate cf" >> ~/.bashrc
SHELL ["/bin/bash", "--login", "-c"]

# Install dependencies
RUN conda install -y pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch-lts -c nvidia

RUN pip install -r requirements.txt 
# Set the working directory to the gridencoder folder
WORKDIR /clip-fields/gridencoder

# Install the hashgrid encoder with the relevant CUDA module
# Update the CUDA_HOME path if necessary
# ENV CUDA_HOME=/usr/local/cuda
# RUN apt-get install -y cuda-runtime-11-1
RUN python setup.py install

# Set the working directory back to the root of the CLIP Fields repository
WORKDIR /clip-fields

# Run any necessary commands or scripts here

# Set the command to run when the container starts
CMD ["/bin/bash"]

它在“> [13/14] RUN python setup.py install”上崩溃，并出现以下错误：

0 0.181 mesg: ttyname failed: Inappropriate ioctl for device                                                                                                                            
#0 0.864 Traceback (most recent call last):                                                                                                                                              
#0 0.864   File "setup.py", line 47, in <module>                                                                                                                                         
#0 0.864     CUDAExtension(
#0 0.864   File "/opt/conda/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 867, in CUDAExtension
#0 0.864     library_dirs += library_paths(cuda=True)
#0 0.864   File "/opt/conda/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 975, in library_paths
#0 0.864     if (not os.path.exists(_join_cuda_home(lib_dir)) and
#0 0.864   File "/opt/conda/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1982, in _join_cuda_home
#0 0.864     raise EnvironmentError('CUDA_HOME environment variable is not set. '
#0 0.864 OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.
------
Dockerfile:35
--------------------
  33 |     # ENV CUDA_HOME=/usr/local/cuda
  34 |     # RUN apt-get install -y cuda-runtime-11-1
  35 | >>> RUN python setup.py install
  36 |     
  37 |     # Set the working directory back to the root of the CLIP Fields repository
--------------------
ERROR: failed to solve: process "/bin/bash --login -c python setup.py install" did not complete successfully: exit code: 1

“FROM pytorch/pytorch:1.8.1-cuda11.1-cudnn8-runtime”是否包含必要的CUDA和pytorch？如何设置 CUDA_HOME？

Answer 1

CUDA_HOME 不能跨层传递。 所以你可能需要在同一行导出 CUDA_HOME 与

python install